> On 21 Mar 2023, at 2:46 pm, Jerome Brunet <jbrunet@xxxxxxxxxxxx> wrote: > > > On Tue 21 Mar 2023 at 11:40, Marc Gonzalez <marc.w.gonzalez@xxxxxxx> wrote: > >> Hello everyone, >> >> I've been benchmarking an Amlogic S905X2 board. >> It provides a BCM4359/9 WiFi chip connected through SDIO. >> >> There's a large performance gap between vendor kernel and mainline. >> (Downloading a 1GB file to /dev/null from a device inches away) > > The title is misleading. You are comparing different sources. This is > not a regression. This is merely a difference. > > If we are talking about mainline, then which board is it ? What is the > corresponding DT ? What is the MMC configuration in both case ? Have > you checked you are running with the same clock configuration to begin > with ? [snip] https://github.com/torvalds/linux/blob/master/arch/arm64/boot/dts/amlogic/meson-g12a-sei510.dts#L428 ^ this device-tree (based on other chat with Marc) To Jerôme’s point; unless you’ve tweaked it upstream will be running at 100MHz and the vendor kernel will be 200MHz, and maybe with other high-speed modes activated. Most upstream device-trees are running a conservative/safe configuration at 50MHz to avoid past speed-related issues. As stability has improved over time most boards could be run at faster speeds, but ‘performance’ is such subjective topic that it would probably need a large amount of testing to convince people to change the current defaults. That said (FWIW) I’ve bumped all boards in my tree that consume the gx-p2xx-q2xx dtsi to 100MHz without any attributable issues being flagged by users. Christian
